Program studiów z ostatnich lat PROGRAM 23 EDYCJI LSNA, 12

Program studiów z ostatnich lat
PROGRAM 23 EDYCJI LSNA, 12-21 września 2012 r.
12-14 września
- Wykład 1: Time Series (15h); Prof. Qiwei Yao (London School of Economics)
Abstract:
Day One:
Examples of time series, objectives of time series analysis
Stationarity, ARMA models, some important nonstationary models
Tests for white noise, tests for random walks
Day Two:
Model identification using ACF, PACF and EACF
Fitting ARMA models: MLE and LSE
Model diagnostics
Model identification based on information criteria
Day Three:
Linear prediction
Modelling heteroscedasticity
ARCH and GARCH models: Basic properties and statistical inference
17-18 września:
- Wykład 2: Statystyczne systemy uczące się ze szczególnym uwzględnieniem metod
uczenia bez nadzoru (10 h); Prof. Jacek Koronacki, Ewa Nowakowska (IPI PAN)
(tytuł angielski: Statistical Learning with Special Emphasis on Unsupervised Learning)
Decyzją Zarządu PSA wykład odbył się w języku polskim
Systemy uczące się, inaczej metody komputerowego uczenia maszynowego (ang.
machine learning), możemy z grubsza utożsamić z komputerowymi metodami
wydobywania wiedzy z obserwowanych danych (ang. data mining). Systemy uczące
się, a wśród nich zwłaszcza metody statystyczne, odgrywają dziś w praktyce bardzo
istotną rolę. Nie ma w tym nic dziwnego – zbieramy ogromne ilości danych i tylko
wykorzystując pamięci i możliwości obliczeniowe komputerów możemy na podstawie
tych danych zdobyć wiedzę o obserwowanym zjawisku.
Wykład rozpoczyna się od omówienia metod uczenia bez nadzoru - rzutowania i
ekstrakcji nowych cech oraz cech ukrytych, i przede wszystkim analizy skupień.
Zasadnicza część wykładu metod rzutowania i ekstrakcji nowych cech oraz cech
ukrytych poświęcona jest skalowaniu wielowymiarowemu, analizie składowych
głównych oraz analizie czynnikowej, a następnie metodom uczenia pod nadzorem –
analizą dyskryminacyjną oraz analizą regresji. Na koniec - metodom uczenia pod
częściowym nadzorem oraz metodom nowym, nieklasycznym.
1
Celem nie jest danie całościowego i teoretycznego wykładu przedmiotu, lecz ujęcie
tematu z perspektywy aktuarialnej.
19-21 września:
- Wykład 3: Mathematical Models and Methods in Life Insurance (15 h); prof. Ragnar
Norberg (Universite Lyon 1)
(Decyzją Komisji Akredytacyjnej PSA zaliczenie wykładu wypełnia wymogi w zakresie
modelowania)
Abstract:
Outline by keywords: A review of classical survival models. Extension to life history
analysis based on time-continuous Markov and semi-Markov models. Statistical
inference. A more general framework of marked point processes and their associated
counting processes and martingales. Application of the models to the analysis of
death benefits, life annuities, health insurance, and more general life insurance
products. Classical life insurance mathematics: the principle of equivalence, reserves,
higher order moments, solvency requirements. Models for environmental risk due to
uncertain development of market indices and demographic indices. Management of
environmental risk: participating policy (or with-profit), index-linked contracts,
securitization through mortality derivatives. Some firm opinions will be articulated,
and discussions are welcome.
2
PROGRAM 22 EDYCJI LSNA, 5-9 września 2011r.
5 – 6 września:
Course 1: Actuarial Modelling for Life Insurance (Frank Schepers, Daniel Matic,
Towers Watson, Niemcy
The course covers introduction to the modelling starting with definitions and
classification of models and also focuses on objectives, selection, calibration and
critical review of models in practice. Examples of practical models and their
applications will give opportunity to understand how crucial modelling is for
insurance business. Models and their components, structure, functionality, areas of
application and relevance in an insurance company will be presented. The course is
dedicated not only for the actuarial students but also wants to bring knowledge in
actuarial modelling required by Associations to become a fully qualified actuary.
7-9 września:
Course 2: Reserves and Reserve Risk in Non-life Insurance (Robert Pusz, PZU SA)
The aim of the lectures is to familiarize participants with reserving and reserve risk in
non-life insurance, in particular with techniques and models worked out in recent
years.
Regarding reserves calculation, participants will learn deterministic methods (Link
Ratios, Chain-Ladder, Bornhuetter-Ferguson, Complementary Loss Ratio) taking into
consideration missing data, outliers, smoothing or extrapolating selected ratios in the
future. Using bootstrap, deterministic methods will be extended to obtain full
distribution of reserves. In order to determine process error, GLM methods will be
presented (Overdispersed Poisson, Mack). The same methods can be bootstrapped
to obtain full reserve distribution. Smoothing data, in particular quarterly data, will be
presented using parametric GLM model with Hoerl curve and its modification as well
as non-parametric GAM model. Next step in the presentation will be using Bayes
methods in reserves calculation. In addition there will be presented influence of
segmentation on reserve calculation and use of correlation matrices and copulas to
model dependencies between lines of business.
Regarding reserve risk, there will be presented Solvency II approach as well as
Swiss Solvency Test approach. Reserve risk will be calculated also by using method
proposed by Merz-Wüthrich. Re-reserving method will be extension and “more
accurate” version of Merz-Wüthrich method i.e. by taking into consideration “tails”.
Lectures will have theoretical as well as practical aspects. Workshops will be
conducted with use of Excel, VBA, R, WinBugs and if needed libraries written in C++,
including using few of above applications in one process.
3
PROGRAM 21 EDYCJI LSNA, 13-17 września 2010 r.
13-15 września:
- Course 1: Beyond Regression: Modern Modeling Techniques with Applications (15
h), Richard A. Derrig PhD, CFE, President, OPAL Consulting LLC, Providence, RI
Visiting Professor Risk, Insurance, and Healthcare Management Fox School of
Business, Temple University, Philadelphia, PA USA, Chair, Committee on the Theory
of Risk, Casualty Actuarial Society
This seminar will cover actuarial methods and applications that use a range of
predictive modeling formats and techniques, both supervised, with an explicit target
variable and unsupervised with a latent target variable. Supervised techniques
presented will include decision trees, neural networks and generalized linear models.
Unsupervised techniques presented will include principal component analysis of ridit
scores and self organizing feature maps. Applications of these techniques will be
selected from both life-health and non-life problems such as risk classification, claim
fraud, medical treatment quality, early duration claims, and equity premiums for risk.
16-17 września:
- Course 2: Replicating Portfolios (10h), Mr. Ateno Villar, PriceWaterhouseCoopers
A replicating portfolio is a hypothetical portfolio of assets selected to closely
match a portfolio of insurance liabilities. If the match is sufficiently close, the
behaviors of the liabilities in a range of scenarios can be examined by
investigating the behavior of the replicating assets portfolio.
The benefit to the insurer is the ease and pace with which the value of
replicating portfolio can be projected forward, relative to the portfolio of
liabilities.
Using replicating portfolios enables the portfolio of liabilities to be valued
much more quickly, using the replicating portfolio as a proxy, following an
important reporting date or significant changes in the financial markets. This is
also applies to complex insurance groups, as replicating portfolios provide an
efficient way to assess the groups’ risk profile, at various confidence levels,
and the group economic capital.
Therefore, replicating portfolios can significantly enhance an insurer’s solvency
and capital management capability.
In addition, representing the portfolio of liabilities as a hypothetical portfolio of
assets can help to develop the understanding of the guarantees and options
embedded into the product design. As a result, replicating portfolios of
liabilities can assist with the communication of the nature of the portfolio of
liabilities by establishing benchmark investment portfolios.
4
Therefore, replicating portfolios can significantly enhance an insurer’s market
risk management.
Abstract:
Day One:
Introduction
Applications in Insurance
Methods for Replicating Portfolio: Balance Sheet method, Aggregate Cash
Flow Method, Cash Flow Matching, Examples
Day Two:
Replicating Portfolios process
Testing Replicating Portfolios
Candidate assets
Replicating Portfolio Summary
5